)]}'
{"doc/source/admin/cpu-topologies.rst":[{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"ffda12cdb2f762229661ff7ac41607c277027594","unresolved":false,"context_lines":[{"line_number":178,"context_line":"By default, instance vCPU processes are not assigned to any particular host"},{"line_number":179,"context_line":"CPU, instead, they float across host CPUs like any other process. This allows"},{"line_number":180,"context_line":"for features like :doc:`overcommitting \u003c/admin/overcommit\u003e` of CPUs. In heavily"},{"line_number":181,"context_line":"contended systems, this provides optimal system performance at the expense of"},{"line_number":182,"context_line":"performance and latency for individual instances."},{"line_number":183,"context_line":""},{"line_number":184,"context_line":"Some workloads require real-time or near real-time behavior, which is not"}],"source_content_type":"text/x-rst","patch_set":1,"id":"bf51134e_53b84c29","line":181,"range":{"start_line":181,"start_character":48,"end_line":181,"end_character":59},"updated":"2020-06-15 17:16:47.000000000","message":"I know you didn\u0027t change this from the previous, but I wonder if it\u0027s worth changing \"performance\" to \"utilization\". Overcommitting your CPU does not improve performance, it improves utilization in times where not all the instances are demanding all the resources they\u0027ve been promised (or allocated). That also might further drive home the point about memory overcommit being purely a utilization hack.","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"}],"doc/source/admin/overcommit.rst":[{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"ffda12cdb2f762229661ff7ac41607c277027594","unresolved":false,"context_lines":[{"line_number":63,"context_line":""},{"line_number":64,"context_line":"CPU overcommit is likely the easiest overcommit ratio to configure as it has"},{"line_number":65,"context_line":"the least chance of resulting in catastrophic failure if misconfigured. CPU"},{"line_number":66,"context_line":"overcommit relies on instance process not running at 100% CPU utilization"},{"line_number":67,"context_line":"continuously."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"CPU resources are represented by the ``VCPU`` resource type in placement and"}],"source_content_type":"text/x-rst","patch_set":1,"id":"bf51134e_33e31836","line":66,"range":{"start_line":66,"start_character":30,"end_line":66,"end_character":37},"updated":"2020-06-15 17:16:47.000000000","message":"I assume you meant \"processes\" or \"an instance process\" or something?\n\nEither way, using the word \"process\" is a bit KVM-specific, even though the same applies to any driver regardless of the mechanism it uses, so maybe just say\n\n \"relies on instances not running at 100% CPU utilization\"\n\n?","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"ffda12cdb2f762229661ff7ac41607c277027594","unresolved":false,"context_lines":[{"line_number":74,"context_line":""},{"line_number":75,"context_line":"In general, CPU overcommit can be configured as high as the load limits of"},{"line_number":76,"context_line":"instances allow. This of course depends highly on your workloads. Setting"},{"line_number":77,"context_line":"overcommit too high will cause the system load average to increase, resulting"},{"line_number":78,"context_line":"in dropped requests or generally unusable response times. You should also avoid"},{"line_number":79,"context_line":"instances overcommitting against themselves, for example, creating an instance"},{"line_number":80,"context_line":"with eight cores on a four core host. Doing so can result in serious"}],"source_content_type":"text/x-rst","patch_set":1,"id":"bf51134e_d359fcfe","line":77,"range":{"start_line":77,"start_character":20,"end_line":77,"end_character":66},"updated":"2020-06-15 17:16:47.000000000","message":"I don\u0027t think this is really the right way to communicate this. First, it\u0027s UNIX specific, so not really relevant for the vmware backend, certainly not Xen (while we have it), and not for hyper-v. Further, a high load average doesn\u0027t necessarily mean that we\u0027re not getting all our work done, nor does it mean that we\u0027re causing the average to increase. Technically any overcommit over 1.0 is promising this, so \"Setting overcommit too high\" also makes it sound like there\u0027s  some value you can set it to, greater than 1.0 that won\u0027t result in the consequences you lay out.\n\nI think I\u0027d just say that setting it too high will cause even instances not demanding 100% CPU to be starved of CPU cycles as there is more competition for those available host CPU cycles.","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"ffda12cdb2f762229661ff7ac41607c277027594","unresolved":false,"context_lines":[{"line_number":86,"context_line":""},{"line_number":87,"context_line":"Memory overcommit, similar to CPU overcommit, relies on the guest not consuming"},{"line_number":88,"context_line":"100% of memory allocated to it. It can be beneficial if correctly configured,"},{"line_number":89,"context_line":"but can result in significant errors if not."},{"line_number":90,"context_line":""},{"line_number":91,"context_line":"Memory resources are represented by the ``MEMORY_MB`` resource type in"},{"line_number":92,"context_line":"placement and can be configured by modifying the ``allocation_ratio`` of the"}],"source_content_type":"text/x-rst","patch_set":1,"id":"bf51134e_137d341c","line":89,"updated":"2020-06-15 17:16:47.000000000","message":"In all cases, at least with KVM, it\u0027s a race against time as any Linux guest that reads $MEMORY_MB data from disk will inflate its own page cache to the size of memory it *thinks* it has. I feel like it might be good to say that in here so people don\u0027t think \"oh well, I want my guests to appear to have 128GB of ram even though it\u0027s just a wordpress instance, and I\u0027ll never use all that.\"","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"},{"author":{"_account_id":2284,"name":"Eoghan Glynn","email":"eglynn@redhat.com","username":"eglynn"},"change_message_id":"f7d7e052540cf39aa3c3b2595fec0d6d885a4b7d","unresolved":false,"context_lines":[{"line_number":92,"context_line":"placement and can be configured by modifying the ``allocation_ratio`` of the"},{"line_number":93,"context_line":"``MEMORY_MB`` resource for a given compute resource provider, whose value is"},{"line_number":94,"context_line":"seeded by the :oslo.config:option:`initial_ram_allocation_ratio` config option,"},{"line_number":95,"context_line":"or by configuring the :oslo.config:option:`ram_allocation_ratio` config option."},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"Memory overcommit is trickier to configure than CPU overcommit and should"},{"line_number":98,"context_line":"generally be kept to a minimum. Swap space is **essential** when using memory"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ff570b3c_d0d36bd2","line":95,"updated":"2020-06-12 08:14:50.000000000","message":"Is it worth noting that the default memory overcommit ratio is (possibly surprisingly) set to 1.5 as opposed to the arguably more intuitive value of 1.0?\n\n(for historical reasons apparently, now lost in the mists of time)","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"},{"author":{"_account_id":2284,"name":"Eoghan Glynn","email":"eglynn@redhat.com","username":"eglynn"},"change_message_id":"f7d7e052540cf39aa3c3b2595fec0d6d885a4b7d","unresolved":false,"context_lines":[{"line_number":100,"context_line":"killer reaping arbitrary processes on the system. A system with a suitably"},{"line_number":101,"context_line":"large swap space will allow for larger memory overcommit ratio, however,"},{"line_number":102,"context_line":"performance may be significantly impacted depending on the workloads and their"},{"line_number":103,"context_line":"memory consumption."},{"line_number":104,"context_line":""},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"Disk Overcommit"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ff570b3c_b05a3780","line":103,"updated":"2020-06-12 08:14:50.000000000","message":"Would it make sense to provide a rule of thumb on what we mean by sufficient swap space?\n\ne.g. circa 4Gb for the host OS plus circa 1Gb per guest for the QEMU process overhead plus the memory overcommit ratio times physical memory?","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"ffda12cdb2f762229661ff7ac41607c277027594","unresolved":false,"context_lines":[{"line_number":100,"context_line":"killer reaping arbitrary processes on the system. A system with a suitably"},{"line_number":101,"context_line":"large swap space will allow for larger memory overcommit ratio, however,"},{"line_number":102,"context_line":"performance may be significantly impacted depending on the workloads and their"},{"line_number":103,"context_line":"memory consumption."},{"line_number":104,"context_line":""},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"Disk Overcommit"}],"source_content_type":"text/x-rst","patch_set":1,"id":"bf51134e_73d75025","line":103,"in_reply_to":"ff570b3c_b05a3780","updated":"2020-06-15 17:16:47.000000000","message":"Yes, we need to state a rule. The rule should be that there should be at least as much swap space as overcommitment. So, if you have 8GB of physical RAM and are overcommitting at 1.5, you need at least 4GB of swap. On Linux especially, you\u0027ll need more than that both for the system itself and the fact that swap accounting itself takes up some memory. I\u0027m not sure that we really need \"GB per guest\" which is also hard to quantify since you don\u0027t typically know how many guests will land on a system. Something like \"enough to cover you overcommitment plus 10%\" might be easy.","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"ffda12cdb2f762229661ff7ac41607c277027594","unresolved":false,"context_lines":[{"line_number":112,"context_line":""},{"line_number":113,"context_line":"Disk overcommit relies on the concept of sparse image files and volumes,"},{"line_number":114,"context_line":"whereby an image or volume uses only the disk space it needs rather than what"},{"line_number":115,"context_line":"has been allocated to it, along with concepts such as thin provisioning."},{"line_number":116,"context_line":""},{"line_number":117,"context_line":"Local disk resources are represented by the ``DISK_GB`` resource type in"},{"line_number":118,"context_line":"placement and can be configured by modifying the ``allocation_ratio`` of the"}],"source_content_type":"text/x-rst","patch_set":1,"id":"bf51134e_d3aa9c83","line":115,"updated":"2020-06-15 17:16:47.000000000","message":"I feel like there needs to be something more here, or in the last paragraph. You don\u0027t want people to think that using sparse files is enough, nor that if they run like this and \"it works for a while\" that everything is good. I know you mention \"continuous monitoring\" in the last paragraph, but it seems like we should mention something about how, in general, even sparse formats _will_ grow to the full size of the virtual disk over time, making any disk overcommit just a bet on how long it will take for that to happen, even with less than 100% guest disk utilization.","commit_id":"56c2791d32cd469c8cd741e9b5f1dac9c0e28a28"}]}